MaxGT

[Request] Midi support

11

Is there any way to send MIDI signals to Pedalboard? Most DAWs can forward MIDI signals to a VST plugin, then [I can use Python as a virtual MIDI device...

Huggingface models are available either in .PT or .H5 format, how would I convert them to a model I can fine-tune with gpt-2-simple? Aitextgen is not an option for me...

I converted my Tensorflow model from gpt-2-simple to Pytorch by running `transformers-cli convert --model_type gpt2 --tf_checkpoint checkpoint/run1 --pytorch_dump_output pytorch --config checkpoint/run1/hparams.json` The resulting "pytorch_model.bin" is completely unable to generate line...

aitextgen and gpt-2-simple trained on same dataset but results not matching AT ALL

1

For simplicity sake, let's say my dataset is structured like this: **`[WP]Elementary, my dear Watson.[RESPONSE]El-e-men-ta-ry, my dear Wat-son.`** With gpt-2-simple, after fine-tuning a pretrained gpt2-medium 3555M model for 1.000.000 steps,...

Is <|startoftext|> broken or deprecated by default?

1

There're open issues that suggests is still being used #88 and #101 . However, does it by default? Because [here](https://github.com/huggingface/transformers/issues/2026#issuecomment-561090890) says otherwise, or it's specifically a feature from aitextgen (such...

Support for CUDA >11.0

2

I tried to make it work with CUDA 11.3, but I learned that anything above CUDA 11.0 isn't supported, is there a way to fix this?